Planning For System Performance - HP StorageWorks P9000 User Manual

Planning for system performance

Remote copy operations can affect I/O performance on the host as well as the main and remote
disk arrays. Continuous Access Synchronous Z provides several options for minimizing the impact
of synchronous operations on performance, or maximize the effectiveness of copy operations to
ensure the best level of backup data integrity. The following options address I/O and performance:
RCU options (see
System- and CU-wide options regarding initial copy (see
concurrently, path watch time" (page
Pair options (see field descriptions in
Your HP service provider can also help you optimize copy operations and performance as follows:
Analyze write-workload. The workload data you collect (MB/s and IOPS) helps determine the
following key elements. When sized properly, they form a data path that operates free of
data bottlenecks under all workload levels. (Bottlenecks severely impact performance.)
◦ Amount of bandwidth
◦ Number of data paths
◦ Number of host-interface paths
◦ Number of ports dedicated for Continuous Access Synchronous Z on the main and remote
system
If you are setting up Continuous Access Synchronous Z for disaster recovery, make sure that
remote disk arrays are attached to a host server. This enables both the reporting of sense
information and the transfer of host failover information. If the remote site is unattended by a
host, you should attach the remote disk arrays to a host server at the main site so that the
system administrator can monitor conditions.
The following lists conditions that affect storage system performance and provides recommendations
for addressing them.
Table 12 Conditions affecting performance, recommendations
Condition
Write-intensive
workloads
Large block size
High host channel
demand
Sequential write
operations
Cache size
30 Planning for Continuous Access Synchronous Z
"RCU maintenance" (page 69)
Description
Write-intensive workloads, such as database
logging volumes, can have a significant impact
on storage system I/O response times.
Workloads with large write block sizes, such
as DB deferred writes, can impact performance.
The demand on the main system host channels
can affect performance.
Continuous Access Synchronous Z operations
can have a negative impact on workloads with
a high percentage of sequential write
operations, such as batch processing operations
(for example, dump/restore, sort operations).
Large cache size improves read performance,
which allows more storage system resources to
be devoted to write operations. Insufficient
cache resources results in command retries,
state-change-pending (SCP) notifications, and
puncture conditions.
and accompanying information)
"Set number of volumes to be copied
53)).
"Create pairs" (page 56))
Recommendation
Spread write-intensive data across several
volumes to minimize queuing.
Spread workloads with large write block sizes
across several volumes.
Spread the workload across several storage
systems to use additional channels.
Avoid performing restore operations to volumes
that belong to Continuous Access Synchronous
Z pairs. Instead, restore data to a scratch
volume, and then create the Continuous Access
Synchronous Z pair.
Consider increasing the cache size of the
Continuous Access Synchronous Z storage
systems to improve overall performance. For
best results, the cache and NVS capacity of
the main and remote system should be the
same. This allows the remote site to function
adequately during disaster recovery.